MODICE tests for likely_ice

I used MOD10A2 (8-day snow composites) for data from August 1-Oct 31 2001 to make "likely ice" masks that we might use for MODSCAG filtering. I turned a mask pixel on if MOD10A2 ever saw snow there for this period. Then I dilated this mask with a 20-pixel (40-pixel) block, which "smeared" out the signal to about 10 km (20 km). See the table below for 10 km (left) and 20 km (right) comparisons.

Then I compared the likely_ice mask with the data we've gotten by processing all MODSCAG pixels (I want to see what we might be missing if we proceed with this method).

We current have 5 tiles processed for both 2001 and 2002 (and I'm working on h25v05/h26v05).

In the table below, I put the himalaya_tiles reference map in the first frame, and then I included each of the tiles we've processed.

Legend: gray=the mask area derived from MOD10A2, blue=MODICE that's inside the mask (so we would still see this), red=MODICE that's outside the mask (so we would miss this MODICE ice if we proceed with the current versions of the mask)

Himalaya Tiles Map
Dilation = ~10km	Dilation = ~20km
h23v04: we need to see if that red blob on the left side is real. the red blob on the right could be caught if we increase the dilation to something bigger than 10 km. (See right hand image--is this enough dilation?
h24v04
h23v05: please click on this one to enlarge. What the heck is that line of MODICE (red) out in the middle of nowhere? Turns out it's a river. Something about turbulence in wide shallow rivers is spectrally similar to snow. So in the case of this river, using the likely ice mask will actualy *improve* the MODICE results.
h24v05: please click on this one, too. Again, there's a MODICE feature in the lower left corner of this image that looks completely erroneous to me. Wha's up with that? --Painter says: another river!
h25v06: This is the Nepal tile. Note that the 10 km dilation is a little too small in a few places. We could increase the dilation to something bigger and this tile would be OK. (See right hand image--is 20 km enough?